The People Behind “Clean” Data
Moderation and labelling work in Manila and Nairobi, and who pays for “safety”
This event is part of a demonstration of what Lera offers for running knowledge events. Its content is written with care and draws on real sources, but the conference has not taken place, and apart from the convener its participants are fictional.
When a conversational assistant declines to describe violence, its behaviour seems natural, as if the courtesy came from the machine itself. But before the machine learned what not to say, people read that violence and gave it names. In January 2023 Billy Perrigo reported in TIME that workers at an outsourcing firm in Nairobi, building a harmful-content filter for one of the best-known of these assistants, spent their days reading and labelling descriptions of sexual abuse, torture and self-harm. Many of them earned less than two dollars an hour.
For more than a decade Manila has been one of the world’s main centres of this work. The 2018 documentary The Cleaners followed young people who every day deleted or kept thousands of images for Western social networks. In Behind the Screen Sarah Roberts called this work “commercial content moderation” and showed that its invisibility is no accident but part of the business model. Mary Gray and Siddharth Suri named it “ghost work”: labour that makes the product possible and appears nowhere in it.
My argument is that the “safety” of AI products is the result of exporting psychological harm. There is a global division of labour at work: those who benefit from a clean model are mostly in the North, and those who clean it are mostly in the South. The bitter irony lies in language. Labelling is done mostly for English, and the workers’ own languages, Tagalog or Swahili, are among the least moderated. The very people who bear the harm of moderation are the least protected online in their own language.
These workers have not stayed silent. In May 2023 more than one hundred and fifty content moderators in Nairobi voted to form a union of African content moderators, and their lawsuits against large companies are still before Kenyan courts. Their organising is a reminder that behind every piece of “clean data” there is organised labour and a claim to rights.
For those who build knowledge systems, two conclusions follow. First, invisible work must become visible in the record of provenance: who labelled a piece of data, where, and under what conditions is part of the history of that knowledge, not a technical detail. Second, moderation in each language should be built by that language’s own community, with fair pay and proper psychological support. Knowledge built on the shoulders of unseen people is not just, however “safe” it may be.
Questions for discussion
- If labellers’ names and working conditions are recorded in data provenance, how is their privacy protected?
- What mechanism could hold the buyers of moderation services answerable for workers’ mental health?
- Do small language communities have the capacity and resources to moderate content in their own language, and if not, who should pay?
References
Perrigo, Billy. 2023. “Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic.” TIME, 18 January 2023.
Roberts, Sarah T. 2019. Behind the Screen: Content Moderation in the Shadows of Social Media. New Haven, CT: Yale University Press.
Gray, Mary L., and Siddharth Suri. 2019. Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Boston: Houghton Mifflin Harcourt.
Block, Hans, and Moritz Riesewieck, dirs. 2018. The Cleaners. Documentary film. Berlin: gebrueder beetz filmproduktion.